Skip to content

Commit 492955e

Browse files
committed
[fix](regression) Bound large Parquet map output batches
Read one row per batch and disable aggregate pushdown for the oversized string-map fixture to retain full decoding with a smaller output allocation.
1 parent 3d78b6b commit 492955e

1 file changed

Lines changed: 4 additions & 1 deletion

File tree

‎regression-test/suites/external_table_p0/tvf/test_hdfs_parquet_group0.groovy‎

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -105,7 +105,10 @@ suite("test_hdfs_parquet_group0","external,hive,tvf,external_docker") {
105105

106106

107107
uri = "${defaultFS}" + "/user/doris/tvf_data/test_hdfs_parquet/group0/large_string_map.brotli.parquet"
108-
order_qt_test_11 """ select count(arr) from HDFS(
108+
// Read both 1 GiB keys one row per batch to avoid a 4 GiB output buffer allocation.
109+
// Disable aggregate pushdown to retain full decoding of the >2 GiB column chunk.
110+
order_qt_test_11 """ select /*+ SET_VAR(batch_size=1, enable_push_down_no_group_agg=false) */
111+
count(arr) from HDFS(
109112
"uri" = "${uri}",
110113
"hadoop.username" = "${hdfsUserName}",
111114
"format" = "parquet"); """

0 commit comments

Comments
 (0)